ARC: Make ARC bitops "safer" (add anti-optimization)

ARCompact/ARCv2 ISA provide that any instructions which deals with
bitpos/count operand ASL, LSL, BSET, BCLR, BMSK .... will only consider
lower 5 bits. i.e. auto-clamp the pos to 0-31.

ARC Linux bitops exploited this fact by NOT explicitly masking out upper
bits for @nr operand in general, saving a bunch of AND/BMSK instructions
in generated code around bitops.

While this micro-optimization has worked well over years it is NOT safe
as shifting a number with a value, greater than native size is
"undefined" per "C" spec.

So as it turns outm EZChip ran into this eventually, in their massive
muti-core SMP build with 64 cpus. There was a test_bit() inside a loop
from 63 to 0 and gcc was weirdly optimizing away the first iteration
(so it was really adhering to standard by implementing undefined behaviour
vs. removing all the iterations which were phony i.e. (1 << [63..32])

| for i = 63 to 0
|    X = ( 1 << i )
|    if X == 0
|       continue

So fix the code to do the explicit masking at the expense of generating
additional instructions. Fortunately, this can be mitigated to a large
extent as gcc has SHIFT_COUNT_TRUNCATED which allows combiner to fold
masking into shift operation itself. It is currently not enabled in ARC
gcc backend, but could be done after a bit of testing.

Fixes STAR 9000866918 ("unsafe "undefined behavior" code in kernel")

Reported-by: Noam Camus <noamc@ezchip.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
This commit is contained in:
Vineet Gupta 2015-07-03 11:26:22 +05:30
parent e2fc61f384
commit 80f420842f

View file

@ -50,8 +50,7 @@ static inline void op##_bit(unsigned long nr, volatile unsigned long *m)\
* done for const @nr, but no code is generated due to gcc \
* const prop. \
*/ \
if (__builtin_constant_p(nr)) \
nr &= 0x1f; \
nr &= 0x1f; \
\
__asm__ __volatile__( \
"1: llock %0, [%1] \n" \
@ -82,8 +81,7 @@ static inline int test_and_##op##_bit(unsigned long nr, volatile unsigned long *
\
m += nr >> 5; \
\
if (__builtin_constant_p(nr)) \
nr &= 0x1f; \
nr &= 0x1f; \
\
/* \
* Explicit full memory barrier needed before/after as \
@ -129,16 +127,13 @@ static inline void op##_bit(unsigned long nr, volatile unsigned long *m)\
unsigned long temp, flags; \
m += nr >> 5; \
\
if (__builtin_constant_p(nr)) \
nr &= 0x1f; \
\
/* \
* spin lock/unlock provide the needed smp_mb() before/after \
*/ \
bitops_lock(flags); \
\
temp = *m; \
*m = temp c_op (1UL << nr); \
*m = temp c_op (1UL << (nr & 0x1f)); \
\
bitops_unlock(flags); \
}
@ -149,17 +144,14 @@ static inline int test_and_##op##_bit(unsigned long nr, volatile unsigned long *
unsigned long old, flags; \
m += nr >> 5; \
\
if (__builtin_constant_p(nr)) \
nr &= 0x1f; \
\
bitops_lock(flags); \
\
old = *m; \
*m = old c_op (1 << nr); \
*m = old c_op (1UL << (nr & 0x1f)); \
\
bitops_unlock(flags); \
\
return (old & (1 << nr)) != 0; \
return (old & (1UL << (nr & 0x1f))) != 0; \
}
#endif /* CONFIG_ARC_HAS_LLSC */
@ -174,11 +166,8 @@ static inline void __##op##_bit(unsigned long nr, volatile unsigned long *m) \
unsigned long temp; \
m += nr >> 5; \
\
if (__builtin_constant_p(nr)) \
nr &= 0x1f; \
\
temp = *m; \
*m = temp c_op (1UL << nr); \
*m = temp c_op (1UL << (nr & 0x1f)); \
}
#define __TEST_N_BIT_OP(op, c_op, asm_op) \
@ -187,13 +176,10 @@ static inline int __test_and_##op##_bit(unsigned long nr, volatile unsigned long
unsigned long old; \
m += nr >> 5; \
\
if (__builtin_constant_p(nr)) \
nr &= 0x1f; \
\
old = *m; \
*m = old c_op (1 << nr); \
*m = old c_op (1UL << (nr & 0x1f)); \
\
return (old & (1 << nr)) != 0; \
return (old & (1UL << (nr & 0x1f))) != 0; \
}
#define BIT_OPS(op, c_op, asm_op) \
@ -224,10 +210,7 @@ test_bit(unsigned int nr, const volatile unsigned long *addr)
addr += nr >> 5;
if (__builtin_constant_p(nr))
nr &= 0x1f;
mask = 1 << nr;
mask = 1UL << (nr & 0x1f);
return ((mask & *addr) != 0);
}